Community Spotlight: Keith Lohse
My name is Keith Lohse and I am a methodologist and applied statistician in Physical Therapy and Neurology at Washington University School of Medicine in Saint Louis. In this role, I primarily work as a co-investigator with other faculty in my department, contributing my expertise in study design, statistical modeling, and machine learning to better understand how people recover following neurological injury.
Tell us about your background
Although I have always been interested in statistics since undergraduate school, my degrees are all in psychology and neuroscience. In time, I drifted from basic cognitive neuroscience into rehabilitation science, but research methods and statistics have always been the unique expertise that I bring to the table. In time, this also led me to develop programming expertise in data management, analysis, and visualization using R and Python.
What sparked your interest in data science, machine learning, and artificial intelligence?
It started with frustration, in many ways. I was increasingly working with complex, longitudinal, and multivariate datasets, which led me to constantly run into new issues, weird exceptions, or simply confusion that forced me to continue to learn new skills. Learning how do these analyses reproducibly in code-based ways also opened up entirely new doors for me in terms of how to manage large datasets, automate otherwise "bespoke" analyses, and collaborate across diverse teams.
What are you most excited about DAPR?
I’m excited to be part of a community that brings together people from different disciplines who are all trying to solve similar problems. Precision rehabilitation requires collaboration between clinicians, engineers, and data scientists, and DAPR creates space for those conversations. I'm especially excited that DAPR gives us the tools to help in a systemic way - we are trying to platform other scientists to collect data that are interoperable, ultimately allowing us to collect massive amounts of data as a community that would be impossible for one team to collect on their own.
What advice would you give to those who wish to follow a similar path?
Focus on fundamentals. Learn principles of statistics, learn a programming language, and develop a deep understanding of your scientific domain. Tools will change, but those foundations will stay with you. Also, don’t feel bad about starting small! Most data are not "big data" and everyone's code is terrible when they are starting out, but if you keep at it, you can really build a powerful and unique knowledgebase in a (relatively) short time.
Is there anything on the horizon of rehabilitation and AI that you’re especially looking forward to?
I’m excited about moving beyond prediction toward explanation, answering "why" questions, and using AI to identify novel mechanisms in recovery. I’m also excited about integrating real-world data from wearables, which can give us a much richer picture of how people actually function in daily life.
Anything you are passionate about outside of DAPR?
Oh, lots of things! Adjacent to DAPR, I really enjoy reading about the philosophy of science and the history of measurement. Farther removed from DAPR, I really love reading, drawing, the outdoors, and hanging out with my wife and our dogs.
Anything else you would like to share? (website/GitHub/podcast)
I’ve combined my interests in science and drawing by creating a series of “Questionable Research Practice Creature Cards”, which illustrate common scientific pitfalls as fantasy monsters. They’re inspired by my lifetime love of Dungeons and Dragons, and more specifically the game HeroQuest that I played as a kid. As I draw them, I post them on my GitHub:
https://github.com/keithlohse/QRP_Creature_Cards
It’s been a fun way to make conversations about rigor and reproducibility more engaging and accessible!
